PCA Truncation On Covariance Matrix - Everyone Talks About It, But What Does It Actually Do?

I bombed 2 quant interviews because I failed PCA questions. Inshallah won't happen again

Welcome back my friends.

Is PCA easy to understand? I first heard about PCA when I was a first year university student. I knew how to write from sklearn.decomposition import PCA. I also watched the best PCA video out there right after it came out.

Steve Brunton is the GOAT

I know if you want to do PCA on covariance matrix, it's just as easy as doing np.linalg.eig(cov). We all know about eigenvalues and eigenvectors right? How hard can it be?

An interview question I was asked before: what does it mean when you run PCA on covariance matrix and one of the eigenvalues is zero? Spoiler: I didn't get the role.

the tweet was directed at myself and one of the candidates I interviewed before (I didn't write that in my CV, but I did write the word "PCA")

It's actually embarrassing to admit how long it took me to feel comfortable with PCA. Even until now I can't say I have "deep understanding" of PCA. But if you look at the portfolio optimisation literature, or factor modelling literature, PCA is everywhere. So in this article, we will try to show what exactly PCA does in context of portfolio optimisation. Hop on, and let's do some linear algebra 😄

No One Goes To That Covariance Town Anymore, It’s Too Noisy

This post is for paying subscribers only

Already have an account? Sign in.

subscribe to quantymacro

sometimes I write cool stuffs